RhinoMCP

RhinoMCP

Control Rhino 3D and Grasshopper with AI, in plain language.

RhinoMCP connects Rhino to AI agents through the Model Context Protocol, so assistants like Claude and Cursor can model geometry, read your document, and build Grasshopper definitions for you, just by chatting.

Quick start · What it can do · Usage · Examples · Tool reference

English · 简体中文

Highlights

For development of the autonomous improvement experiments, start with the continuation guide.

• Describe what you want and the assistant builds it in Rhino.
• It reads your document and can capture the viewport, so it works from what is actually on screen.
• It scripts Grasshopper for you: finding components, wiring them, setting sliders, and solving.
• A single plugin and a single config entry cover both Rhino and Grasshopper.
• When you need more control, it can run native Rhino commands, RhinoScript-Python, or RhinoCommon C#.

[!NOTE] RhinoMCP targets Rhino 8 on Windows and macOS.

Demos

Prefer a walkthrough? Nate made a showcase and install tutorial on YouTube.

What it can do

Rhino

Grasshopper

Quick start

Three steps: install the Rhino plugin, connect your AI client, then start the bridge in Rhino.

1. Install the Rhino plugin

In Rhino, open Tools → Package Manager, search for rhinomcp, and click Install. Restart Rhino.

2. Connect your AI client

Option A: ask your AI assistant to install it (recommended)

If you use an agentic assistant (Codex, Claude Code, Cursor, Cline, and the like), paste this prompt:

Please install https://github.com/jingcheng-chen/rhinomcp as a local MCP server named `rhino`.

Option B: Install the mcp server or manually edit the config yourself

Codex, in one command:

codex mcp add rhino --env RHINO_MCP_HOST=127.0.0.1 -- uvx rhinomcp@latest

Claude Code, in one command:

claude mcp add rhino -- uvx rhinomcp@latest

ChatGPT: use Codex for the local setup above. ChatGPT apps/MCP connectors currently connect to remote MCP servers, not local stdio commands like uvx rhinomcp. If you want to build a ChatGPT app around RhinoMCP, use ChatGPT developer mode with a remote or tunneled MCP endpoint.

You can also manually edit the config yourself:

{
  "mcpServers": {
    "rhino": {
      "command": "uvx",
      "args": ["rhinomcp@latest"],
      "env": {
        "RHINO_MCP_HOST": "127.0.0.1"
      }
    }
  }
}

[!IMPORTANT] The launcher uvx comes from uv. If you don't have it yet: macOS brew install uv · Windows powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

Run only one RhinoMCP server at a time (Codex, Claude, Cursor, etc. — not several at once).

{
  "mcpServers": {
    "rhino": {
      "command": "sh",
      "args": ["-c", "killall rhinomcp 2>/dev/null; uvx rhinomcp@latest"]
    }
  }
}

{
  "mcpServers": {
    "rhino": {
      "command": "cmd",
      "args": ["/c", "taskkill /F /IM rhinomcp.exe 2>nul & uvx rhinomcp@latest"]
    }
  }
}

3. Start the Rhino bridge

With Rhino open, type mcpstart in the command line. This starts the TCP bridge the server connects to (mcpstop ends it). Run it once per Rhino session.

Staying up to date

RhinoMCP has three parts that update separately:

• Python server. uvx rhinomcp@latest asks PyPI for the newest release every time your client starts the server; plain uvx rhinomcp keeps the first version it downloaded. If you work offline, install it once with uv tool install rhinomcp and run uv tool upgrade rhinomcp when convenient.
• Rhino plugin. Installed through the Package Manager, Rhino checks for updates at startup and installs them when "Automatically update packages when Rhino starts" is enabled; the new version loads on the next restart. A plugin installed by dragging a .rhp into Rhino is not updated.
• AI client. It reads the tool list when it launches the server, so restart the client (or reconnect the server) after an update.

The server reads the plugin's version when it connects and warns in its log when the two differ. A tool that needs a newer plugin fails with an update instruction instead of silently doing something else, and describe_capabilities reports both versions.

Usage

With the bridge running and your client connected, you'll see the RhinoMCP tools. From there, just chat: ask the assistant to model something, inspect your scene, or build a Grasshopper graph.

For Grasshopper, you only need Rhino open with mcpstart running. The assistant can open or create the Grasshopper document itself, then build the definition. For example: "create a point attractor pattern with cylinders that have different heights."

Example prompts

Create 6×6×6 boxes on a 10-unit grid from the origin, sizes ramping from 1 to 5, with a blue-to-red gradient color based on size. Use RhinoScript Python.

Make a Rhinoceros animal out of cubic blocks in cartoon colors. Then change its head to red, and rotate the selected object 90° around the Z axis.

Create a point attractor pattern in Grasshopper: a grid of cylinders whose heights change with their distance from an attractor point, with a slider to move the point.

Tool reference

How it works

AI client ──MCP (stdio)──► rhinomcp (Python) ──TCP 127.0.0.1:1999──► Rhino plugin ──► Rhino + Grasshopper

1. server/: a Python MCP server (official Python SDK 2.x) that exposes each tool and forwards it to Rhino.
2. plugin/: a RhinoCommon C# plugin that runs a TCP listener inside Rhino and executes commands on the main thread. Start and stop it with the mcpstart / mcpstop Rhino commands.
3. contracts/: JSON Schema definitions that keep the wire protocol between the two tiers in sync.

IMPLEMENTATION.md has a deeper tour of the code.

Security

The Python server and the Rhino plugin talk over an unauthenticated TCP loopback link (127.0.0.1:1999). Tools such as run_command, execute_rhinoscript_python_code, and execute_rhinocommon_csharp_code give the model an open execution surface inside Rhino. This is fine for local agent use. Do not expose it beyond the loopback interface without adding authentication.

For developers

uv venv && uv pip install -e ".[dev]"   # set up
uv run pytest                            # run tests (no Rhino needed; uses a mock server)
uv run ruff check src/rhinomcp           # lint
uv run python ../contracts/test_schemas.py   # validate JSON schemas
uv build && uv publish                   # publish to PyPI

dotnet restore plugin/rhinomcp.sln
dotnet build plugin/rhinomcp.sln --configuration Release

Contributing

Contributions are welcome. Feel free to open an issue or submit a pull request.

Disclaimer

This is a third-party integration and is not made by McNeel. Built by Jingcheng Chen.

Star history

Modeling guidance distributed with the server

The Python package includes reusable guidance for transforms, planar regions, organization, verification and recovery. Use get_modeling_guidance("overview"), the asset_general_strategy MCP prompt, or the rhinomcp://guidance/{topic} resource. Essential tool semantics stay in tool descriptions. No local experiment folder or separate skill is needed. See packaged guidance for available topics and release details.